DOJO – A SuperComputer designed by Tesla
Tesla described its incredible lengths in building Dojo, its exaflop class supercomputer. It also throws a light on what it will be capable of in terms of data processing. This amazing feature is unquestionably going to place Tesla in the higher spot and perhaps ahead of the most powerful computers in the world. Once completed, it is predicted to be the fastest computer in the world for AI training.
According to Tesla, the major reason behind designing their own supercomputer was that it found existing solutions insufficient for the processing power and volume of data that is required to propel its Full Self-Driving (FSD) training needs. Tesla has carried FSD to a usable level, yet it needs Dojo to drive it further.
D1 chip, the main element of the Dojo supercomputer, was designed first for maximum bandwidth, laying in communication channels and other neural processing components. 25 of these D1 chips are spread out in a five-by-five framework to make a preparation tile with 9 petaflops of execution.
Tesla guarantees multiple times the performance and around one quarter less energy when compared to existing systems. With this amount of AI preparing power, the FSD framework will actually improve more quickly and handle uncommon driving situations all the more smoothly.
As claimed, it is an AI-specific computer designed to enhance machine learning and various other learning activities. Its lesser focus limits the application to a broader HPC context.
The competencies mentioned by Dojo do not supply its genuine high-performance computer (HPC) status,
Dojo supercomputer aims to be the industry topper in machine learning training. Having said that, calling it a supercomputer and claiming it will break the exaflop barrier, can be difficult to promote while everyone else is rating their systems using standards that are twice as complicated.
Dojo is Powered by Tesla’s custom chip D1, which contains:
- Technology: 7nm and
- Processing power: 362 teraflops
- GPU level compute and CPU level flexibility
- Chips can connect seamlessly
- IO networking switch
- Needs 120 fully functional wafers
Following are the competitors for Dojo Supercomputer:
- Chipmaker Intel
- Graphics card maker Nvidia, and
Tesla has a history of coming up with extraordinary concepts that never materialized. Many of the products hyped up during the events weren’t commercialized, aren’t available yet, or were eliminated at the ideation level. They haven’t published any official research paper on the supercomputer either.
You may also like Tesla EVs banned from China government compounds